Attributing Snowpack Biases over the Contiguous U.S. in Four U.S. CMIP6 Models to Temperature and Precipitation Biases

Snow is an important feature of the wintertime land. It changes the surface energy balance by increasing surface albedos and insulating the ground from the atmosphere. It also changes the surface water balance by storing water that is released later during snowmelt. Therefore, its representation in weather and climate models is critical. Recently, we developed the first gridded snow water equivalent (SWE) and snow depth dataset at 4-km resolution over the contiguous U.S. (CONUS) from 1981 to present derived from upscaling in situ snowpack measurements. We use this dataset here to evaluate snowpack simulated by the Atmospheric Model Intercomparison Project simulations from four U.S. contributions to CMIP6 (E3SMv1, CESM2, GFDL CM4, and GISS ModelE). SWE is generally underestimated over the western CONUS, while it is generally overestimated over the northern CONUS. We will attribute these biases to model biases in temperature and precipitation relative to the PRISM dataset or to model formulation.

Furthermore, model trends and interannual variability in SWE will be evaluated and attributed to biases in temperature and precipitation. For instance, trends in wintertime (October-March) temperature in most models is a dipole with warming in the eastern CONUS and cooling in the western CONUS. This contrasts with the PRISM trends of warming everywhere with the most warming in the West. We will show how these trends along with those in wintertime accumulated precipitation contribute to biases in the simulated SWE trends in both regions. Maximum SWE from the 4-km dataset is strongly correlated to winter mean temperatures and accumulated precipitation in PRISM, but E3SM and CESM2 have weaker correlations to these. Such correlations are stronger in the GFDL and GISS models.